Does my voice ever leave my Mac?

No. Dictation and transcription run on-device, so your audio stays on your machine. The only outbound request is the optional AI cleanup, which sends text, not audio, to the provider whose key you supply.

Does WhisprDesk collect telemetry or require an account?

No. The app has no account, no sign-in, and no analytics or telemetry.

Is there any telemetry or usage tracking in the app?

No. The application contains no telemetry and no usage analytics, and there is no account to attach activity to. Note this is a statement about the Mac app; the marketing website, like most sites, uses a privacy-first cookieless visit counter, which is described on our privacy page.

Can I run it on a machine with no network access at all?

Not as the intended setup. Transcription itself needs no connection once your model has downloaded, but a licensed copy re-checks its licence periodically. Ordinary offline use works; a permanently air-isolated machine is not what this is built for.

Is it suitable for confidential material?

On-device processing means your audio is not uploaded and there is no server-side copy, which is why it suits sensitive notes and confidential work. You remain responsible for how you store the resulting text.

Private dictation for Mac

Private dictation means your voice is turned into text on your own device instead of on a company's servers. WhisprDesk is built this way: live dictation and audio transcription run on your Mac using Whisper or NVIDIA Parakeet, so the audio never leaves your machine. There is no account to create and no telemetry sent back, which removes the usual places a dictation tool could collect or leak what you say. That matters for anyone who dictates sensitive material, from personal notes to work that is under confidentiality. Optional AI cleanup is opt-in and uses your own API key, so even that step stays under your control. WhisprDesk runs on Apple Silicon, macOS 12 and later, and is available now with a 7-day free trial.

What “private” actually means here

Privacy claims are easy to make and hard to verify, so it helps to be concrete. WhisprDesk processes your speech on-device: the model runs on your Mac, the audio is not uploaded, there is no account, and the app sends no telemetry. There is no server-side copy of your voice because your voice never goes to a server.

What WhisprDesk does not collect

The WhisprDesk app has no sign-in, and no analytics or telemetry. It does not create an account, does not phone home with usage data, and does not upload your audio. The only outbound request is one you trigger: the optional AI cleanup, which sends text (not audio) directly to the AI provider whose key you supply.

Live dictation and file transcription, both local

Both ways you use your voice stay on the device. Live, system-wide dictation lets you speak into the app you are using, and audio-file transcription turns a recording into text. Both run locally with Whisper or NVIDIA Parakeet.

The three things that could send data, and what each does

Rather than ask you to trust a privacy promise, it is more useful to enumerate every path out of the machine, because there are only three and two of them are off unless you turn them on.

Transcription itself. By default this runs on a model stored on your Mac, so nothing is transmitted. The app can be switched to use a cloud model with your own API key, and that is the only configuration where your audio leaves. It is off by default.

AI cleanup. If you ask a model to tidy a transcript, the text of that transcript goes from your Mac directly to the provider whose key you supplied. Your audio is not sent, and no server of ours sits in between. This is off unless you configure a key.

Licence checking. A licensed copy contacts the licensing service periodically to confirm it is still valid. That carries no audio and no transcript. It is worth stating plainly because it means a permanently disconnected machine is not the intended setup, even though ordinary offline work is fine.

What the app does not do

There is no account and no sign-in, so there is no profile to correlate activity against. There is no telemetry or usage analytics in the application.

Where the app runs a local service for integrations, it binds to your loopback interface only, so it is reachable from your own Mac and not from the network around it.

Speech models are fetched once from their source and cached on your disk under ~/.cache/whisprdesk/. After that, choosing or switching models involves no further network calls.

Optional AI cleanup stays opt-in

If you want an AI model to tidy or reformat the text, that is a choice you make per use, with your own OpenAI, Anthropic, or Google key. It is never required, and the dictation itself never depends on it.

Frequently asked questions

Does my voice ever leave my Mac?: No. Dictation and transcription run on-device, so your audio stays on your machine. The only outbound request is the optional AI cleanup, which sends text, not audio, to the provider whose key you supply.
Does WhisprDesk collect telemetry or require an account?: No. The app has no account, no sign-in, and no analytics or telemetry.
Is there any telemetry or usage tracking in the app?: No. The application contains no telemetry and no usage analytics, and there is no account to attach activity to. Note this is a statement about the Mac app; the marketing website, like most sites, uses a privacy-first cookieless visit counter, which is described on our privacy page.
Can I run it on a machine with no network access at all?: Not as the intended setup. Transcription itself needs no connection once your model has downloaded, but a licensed copy re-checks its licence periodically. Ordinary offline use works; a permanently air-isolated machine is not what this is built for.
Is it suitable for confidential material?: On-device processing means your audio is not uploaded and there is no server-side copy, which is why it suits sensitive notes and confidential work. You remain responsible for how you store the resulting text.